AI Investment Faces ‘Low Point of Disillusionment’ as ROI Concerns Mount
Table of Contents
Despite the rapid advancement of generative AI, a staggering 95% of corporate projects fail to deliver a return on investment, signaling a critical need to refocus strategies on process innovation and robust governance. This was the central message delivered at the ‘AX & Hyperautomation korea 2025’ conference by Byeong-seop, Head of AI at Shinhan Financial Group.
The AI landscape has been described as “frightening,” with expectations for generative AI peaking earlier this year before falling sharply, according to Byeong-seop.he cautioned against the common pitfall of simply introducing technology without a clear understanding of how it integrates with existing workflows. “the approach of expecting results just by introducing technology is fundamentally wrong,” he stated. Many companies, he explained, are caught in a cycle of starting AI initiatives based on pressure to adopt the technology, only to stall at the proof-of-concept stage or face governance issues after launch.
A key challenge lies in the inherent limitations of generative AI for corporate automation. While powerful,the technology is not flawless and requires meaningful human oversight. “The irony is that in order to use AI well, you have to sit in front of the PC for longer,” Byeong-seop noted, highlighting the need for continuous correction and refinement.
The Power of AI Agents: Beyond Sentence Generation
Byeong-seop showcased a prosperous implementation of AI agents with the ‘Loan Agreement AI Agent,’ wich automates the processing of over 60,000 loan agreements monthly. This agent leverages a workflow that minimizes human intervention, incorporating document reception, semantic understanding via Large Language Models (LLMs), data extraction, AI governance verification, structured data generation, and integration with Robotic Process Automation (RPA).
The core of a successful enterprise AI agent, he emphasized, is “the combination of LLM and AI governance, and the provision of structured data.” He argued that generative AI’s tendency to provide answers in natural language is insufficient for business process automation,advocating for an LLM role focused on ‘semantic-based extraction’ rather than ‘sentence generation.’
AI as a Resource, Not Just a Tool
Metanet Global CEO kim Ki-ho, joining Byeong-seop for a discussion, articulated a shift in outlook on AI. He explained that AI is evolving from a mere ‘tool’ to a valuable ‘resource’ capable of independently generating value.Metanet Global has responded by establishing an AI studio to facilitate rapid agent creation, resulting in productivity gains through automation of tasks like insurance reporting and year-end tax settlements.
Ki-ho’s vision extends to a fundamental reorganization of the corporate operating model, stating, “We will create humanless services, and our employees will become an R&D organization that produces those services.” He stressed the importance of defining clear roles and evaluation systems for AI agents, treating them akin to employees within the organization.
Three Pillars of Successful AI Agent Implementation
Byeong-seop concluded his presentation by outlining three critical conditions for successful AI agent implementation:
- High-Quality Data: Given that much corporate data resides in documents,a combination of RPA,Clever Document Processing (IDP),and LLMs is essential for handling structured,semi-structured,and unstructured data.
- Robust Governance: Data sources, verification processes, audit trails, and approval procedures are paramount for process-based AI agents.
- Process Integration: ROI is directly tied to how seamlessly AI technology integrates with existing and subsequent processes, emphasizing that full automation is an evolutionary process, not a revolution.
“For a company’s AI agent project to succeed, data, governance, and the design of an automated process without human intervention are more significant than the technology itself,” Byeong-seop reiterated, underscoring the need for a holistic and strategic approach to AI adoption.
